Optimizing Mendelian Randomization for Drug Prediction: Exploring Validity and Research Strategies

Abstract Mendelian randomization (MR) plays an increasingly important role in drug discovery, yet its full potential and optimized framework for accurately predicting drug targets have not been firmly established. This study aimed to evaluate the efficacy of multiple MR models in predicting effective drug targets and to propose the optimal selection of models and instrumental variables for MR analyses. We meticulously constructed datasets using approved drug indications and a range of IVs, encompassing cis-expression quantitative trait loci (eQTLs) and protein quantitative trait loci (pQTLs). Our analytical approach incorporated diverse models, including Wald’s ratio, inverse-variance weighted (IVW), MR‒Egger, weighted median, and MRPRESSO, to evaluate MR's validity in drug target identification. The findings highlight MR efficacy, demonstrating approximately 70% accuracy in predicting effective drug targets. For the selection of instrumental variables, tissue-specific eQTLs in disease-related tissues emerged as superior IVs. We identified a r 2 threshold below 0.3 as optimal for excluding redundant SNPs. To optimize the MR model, we recommend IVW as the primary computational model, complemented by the weighted median and MRPRESSO for robust analyses. This finding is consistent with current findings in the literature. Notably, a P value of < 0.05, without false discovery rate correction, is the most effective for identifying significant drug targets. With the optimal strategies we summarized, we identified new potential therapeutic targets for IBD and its subtypes, including ERAP1, HLA-DQA1, IRF5 and other genes. This study provides a refined, optimized strategy for MR application in drug discovery. Our insights into the selection of instrumental variables, model preferences, and parameter thresholds significantly enhance MR's predictive capacity, offering a comprehensive guide for future drug development research.